LightRAG: Graph-Powered Analysis for Portfolio Company Relationships

At Zen Trading, we're constantly pushing the boundaries of AI-powered investment research. Today, we're excited to share how we've integrated LightRAG—a cutting-edge graph-based Retrieval-Augmented Generation framework—to analyze complex relationships between our portfolio companies.

Traditional RAG systems rely on flat data representations, limiting their ability to understand intricate relationships between entities. LightRAG changes this by incorporating knowledge graphs into text indexing and retrieval, enabling us to capture the full context of interconnected portfolio companies.

Why LightRAG for Portfolio Analysis?

When analyzing portfolio companies, investors often face questions like: "How does Company A's supply chain exposure affect Company B's revenue?" Traditional RAG systems might retrieve separate documents on each company but struggle to synthesize this information into coherent insights that capture complex inter-dependencies.

LightRAG addresses these challenges through its innovative dual-level retrieval paradigm. Low-Level Retrieval focuses on specific entities—individual companies, executives, products—and their direct relationships. High-Level Retrieval captures broader themes—market trends, sector dynamics, competitive landscapes.

Our Implementation

We construct our portfolio knowledge graph by ingesting SEC filings, earnings transcripts, news articles, and analyst reports for each portfolio company. We use LLMs to identify companies, executives, products, partnerships, and financial metrics, then establish connections like supplies_to, competes_with, partners_with, and invests_in.

Key Benefits

LightRAG's graph structure enables extraction of global information from multi-hop subgraphs, helping us understand how a disruption at one company ripples through our entire portfolio. Compared to GraphRAG's community traversal approach, LightRAG uses fewer than 100 tokens for retrieval versus 610,000+ tokens, dramatically reducing our inference costs.

Conclusion

LightRAG has transformed how we analyze portfolio company relationships at Zen Trading. By leveraging knowledge graphs and dual-level retrieval, we can uncover hidden connections between seemingly unrelated holdings and synthesize cross-company insights that traditional RAG systems miss.

References

Guo, Z., Xia, L., Yu, Y., Ao, T., & Huang, C. (2024). LightRAG: Simple and Fast Retrieval-Augmented Generation. arXiv:2410.05779

Paper: arXiv:2410.05779
Repository: LightRAG GitHub Repository